Analysis of Trade-offs in Fault-Tolerant Distributed Computing and Replicated Databases

This paper examines fundamental trade-offs in fault-tolerant distributed systems and replicated databases built over the Internet. We discuss interplays between consistency, availability, and latency which are in the very nature of globally distributed computer systems and also analyse their interconnection with durability and energy efficiency. In this paper we put forward an idea that consistency, availability, latency, durability and other properties need to be viewed as more continuous than binary in contrast to the well-known CAP/PACELC theorems. We compare different consistency models and highlight the role of the application timeout, replication factor and other settings that essentially determine the interplay between above properties. Our findings may be of interest to software engineers and system architects who develop Internet-scale distributed computer systems and cloud solutions.